"what" meaning "the thing that"
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 19 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron fires on the word "what" when it means "the thing that", as in "into what was a traditional product". Its weak negative side is on words like "a" and "your". Turning it up made the text repeat hedging words like "often" and "probably".
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
Scripture which, when you read them, seem out of place with the passages that surround it. It makes you wonder why this particular Bible writer included this portion in what he was writing. You may remember when we were studying the Book of Judges,
-
acillus bulgaricus) that ferments milk into yogurt. Grigorov’s study paved the way for the commercialization of yogurt, breathing new life into what was until then a traditional product that people made almost exclusively at home.⏎Satisf
-
the moment Parks started a revolution of conscience in 1955 was more than eager to pay its respects. More than eager to recall a proud moment in what we once felt was our collective history. From social crusaders to die-hard segregation
-
produced as a result of the process.⏎Dirty Diapers Turned into Viable Diesel: one company has a brilliant (albeit offbeat) idea for turning what would normally be landfill material into cost-effective, synthetic diesel fuel. Approximately 3
-
a powerful medicine for good. These boys learned something—about their own home city (for they live in Brooklyn, and the battle they investigated had been fought along what are now the familiar streets and squares and parks of their own neighborhood), about the
-
s Route? Ker Than, for National Geographic News, Published November 1, 2011 Under a former Native American village in Georgia, deep inside what’s now the U.S., archaeologists say they’ve found 16th
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: the idea that a day on the public domain is a day on the public domain. What about the public domain does this idea do a day on the public domain on the public domain Type the day on the public domain on the public domain
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
moment, but symptoms will slowly fade over time as a person’s hormones return to normal.⏎Addiction changes the chemical structure and function of
-
Dehydration can make you more vulnerable to illness and infections, therefore, it is important that you down plenty of non-alcoholic fluids. And if water isn’t your thing, juice, tea, and other beverages
-
in between the layers.⏎It comes in many forms and consistencies depending on the amount of ingredients and how long you cook it for. However, there is such a thing as the perfect sauce
-
Levy, the son of Polish-immigrant parents and a prominent Charleston merchant, and the Liverpool-born Fanny Yates, a successful Charleston actress. Phoebe was the fourth of their seven children.⏎The family was strongly loyal to the Confederate cause.
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L10:1774 weakly suppresses
- L17:1833 weakly suppresses
- L2:20 weakly suppresses
- L1:903 weakly suppresses
- L7:22 weakly suppresses
- L13:460 weakly suppresses
- L9:1081 weakly excites
- L9:1085 weakly suppresses
- L15:2873 weakly suppresses
- L8:1331 weakly excites
- L14:2402 weakly suppresses
- L5:11 weakly excites
- L15:186 weakly suppresses
- L2:819 weakly excites
- L10:1604 weakly suppresses
- L15:2229 weakly suppresses
- L10:1124 weakly suppresses
- L4:137 weakly excites
- L14:2472 weakly excites
- L18:313 weakly excites
- L10:2127 weakly excites
- L18:1035 weakly excites
- L18:1084 weakly excites
- L11:1278 weakly suppresses
Feeds later layers
- L24:291 excites
- L23:377 excites
- L20:2909 excites
- L21:2227 excites
- L25:1767 excites
- L27:644 excites
- L21:1245 excites
- L22:2556 excites
- L21:414 weakly excites
- L21:1396 weakly excites
- L21:1362 weakly suppresses
- L20:2104 weakly excites
- L20:1010 weakly excites
- L25:1264 weakly excites
- L22:2721 weakly excites
- L25:2039 weakly excites
- L21:1983 weakly excites
- L23:485 weakly excites
- L26:1554 weakly excites
- L20:302 weakly suppresses
- L26:1848 weakly excites
- L24:1043 weakly excites
- L21:607 weakly excites
- L24:2686 weakly suppresses
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- presumably
- often
- otherwise
- mostly
- usually
- probably
- mostly
- typically
- largely
- essentially
Suppresses
- 纵
- .Std
- спект
- tử
- .IsMatch
- động
- VD
- ingham
- ジャー
- 黻